mirror of
https://github.com/lilakk/BLEUBERI.git
synced 2026-04-19 12:58:12 +00:00
37 lines
No EOL
1.2 KiB
Bash
37 lines
No EOL
1.2 KiB
Bash
model_name="microsoft/Phi-3-mini-128k-instruct"
|
|
model_pretty_name="Phi-3-mini-128k-instruct"
|
|
TEMP=0; TOP_P=1.0; MAX_TOKENS=4096;
|
|
batch_size=4;
|
|
# gpu="0,1,2,3"; num_gpus=4;
|
|
|
|
CACHE_DIR=${HF_HOME:-"default"}
|
|
output_dir="result_dirs/wild_bench_v2/"
|
|
|
|
|
|
# Data-parallellism
|
|
start_gpu=0
|
|
num_gpus=1
|
|
n_shards=4
|
|
shard_size=256
|
|
shards_dir="${output_dir}/tmp_${model_pretty_name}"
|
|
for ((start = 0, end = (($shard_size)), gpu = $start_gpu; gpu < $n_shards+$start_gpu; start += $shard_size, end += $shard_size, gpu++)); do
|
|
|
|
CUDA_VISIBLE_DEVICES=$gpu \
|
|
python src/unified_infer.py \
|
|
--start_index $start --end_index $end \
|
|
--data_name wild_bench \
|
|
--model_name $model_name \
|
|
--max_model_len 40960 \
|
|
--use_hf_conv_template \
|
|
--download_dir $CACHE_DIR \
|
|
--tensor_parallel_size $num_gpus \
|
|
--dtype bfloat16 \
|
|
--model_pretty_name $model_pretty_name \
|
|
--top_p $TOP_P --temperature $TEMP \
|
|
--batch_size $batch_size --max_tokens $MAX_TOKENS \
|
|
--output_folder $shards_dir/ \
|
|
--overwrite &
|
|
done
|
|
wait
|
|
python src/merge_results.py $shards_dir/ $model_pretty_name
|
|
cp $shards_dir/${model_pretty_name}.json $output_dir/${model_pretty_name}.json |